About  ·   Work  ·   Research  ·   Stack  ·   Analytics  ·   Contact

Research paper  ·   Brain2Text  ·   LinkedIn

I build AI systems end to end: from data and model behavior to APIs, interfaces, and deployment. My work spans neural speech decoding, multi-agent reinforcement learning, computer vision, and full-stack product engineering.

The thread connecting it all is practical research—systems should be reproducible, observable, and usable beyond the notebook. I am especially interested in machine learning systems, deployment engineering, and human-centered AI.

• Co-author of Brain2Text, an open-access 2026 research article on CPU-deployable neural speech decoding.
• Builder of public projects across reinforcement learning, assistive computer vision, and simulation.
• AWS Certified Cloud Practitioner, with hands-on work across Python, PyTorch, React, Flask, Docker, Supabase, and PostgreSQL.

An end-to-end neural decoding pipeline that maps 512-dimensional intracortical features to phonemes with a five-layer GRU + CTC decoder, then reconstructs English text. The project packages CPU inference behind a Flask API with a React/Vite interface and an offline demonstration mode.

PyTorch GRU / CTC Flask React Neural decoding

A self-learning multi-agent football environment built around PPO and competitive self-play. It combines reward engineering, position-neutral training, persistent scoring, and real-time Pygame visualization so learning behavior can be inspected as it evolves.

Python PPO Stable-Baselines3 PettingZoo Pygame

A real-time assistive computer-vision system that turns webcam hand landmarks into alphabet predictions. MediaPipe handles landmark extraction; a machine-learning classifier performs gesture recognition with live visual feedback.

OpenCV MediaPipe scikit-learn Computer vision Real-time ML

A deployed last-mile delivery simulator with dynamic package assignment, an interactive Canvas visualization, and per-agent efficiency analytics. It exposes routing logic through a Flask backend and supports JSON input plus CSV export.

Flask JavaScript HTML Canvas Simulation Analytics

Gaurav Kumar Singh, Aayush Chougule, Uday Tomar, and Sidheshwar Sharma. “Brain2Text: A Reproducible, CPU-Deployable Framework for Neural Speech Decoding with Browser-Accessible Inference.” International Research Journal of Innovations in Engineering and Technology, 10(5), 618–624, 2026. DOI · Full text · Code

• Reproducible AI demos — model weights, inference paths, setup notes, and interfaces packaged together where possible.
• Learning in the open — public implementations for neural decoding, reinforcement learning, computer vision, and simulation.
• Contributor-friendly foundations — documented setup and contribution paths in projects such as Sign Language Detection.

_{The analytics card and contribution animation are generated inside this repository by GitHub Actions. GitHub’s native contribution calendar and activity feed continue immediately below the profile README.}

I’m interested in AI/ML engineering, research collaboration, and systems that turn promising models into dependable products.

Connect on LinkedIn · Explore my GitHub · Read Brain2Text